Korea’s AI Basic Act (인공지능 기본법, in full 인공지능 발전과 신뢰 기반 조성 등에 관한 기본법) took full effect on January 22, 2026. Most enforcement is on hold under a one-year grace period (계도기간, gyedo-gigan, “guidance period”) — but two rules are not on hold. Missing an AI notice or a domestic representative is already fined, right now, up to ₩20 million.
I found this out the annoying way: by telling a coworker the wrong thing, then having to walk it back a week later.
What I Told a Coworker That Turned Out to Be Wrong
A few weeks ago, one of the non-Korean developers on my team asked me about the AI Basic Act over lunch. He’d seen a headline about it and wanted to know if it meant anything for us — I work IT at a Korean financial company, the kind of place that runs loan-screening models and internal HR tools on a daily basis. I told him, with more confidence than I’d earned, that there was a one-year grace period, so realistically nothing would happen until early 2027. Relax, I said. Compliance will sort it out.
That was only half true, and the half I got wrong turned out to be the part that actually applies to us. Our compliance team sent around an internal notice the following week clarifying that two specific violations are enforced starting immediately — not after the grace period ends. I’d conflated “most things are deferred” with “everything is deferred,” which is exactly the mistake the law is set up to catch companies making.
What “High-Impact AI” (고영향 인공지능) Actually Covers
The AI Basic Act doesn’t regulate every algorithm in the country the same way. It draws a line around “high-impact AI” (고영향 인공지능, goyeonghyang injigeuneung, literally “high-influence artificial intelligence”) — systems that can meaningfully affect a person’s life, safety, or basic rights. The law names roughly ten sectors where this applies: healthcare, energy, transportation, public services, and — the one that put this law on my desk specifically — hiring and loan/credit screening (채용·대출심사).
That last pair is why my employer is squarely inside the law’s scope rather than reading about it as a bystander. A Korean bank that scores loan applications with a model, or ranks job candidates with a resume-screening tool, is running exactly the kind of system the law was written to catch. Our credit-risk models and our recruiting platform’s résumé filter both fall under “high-impact AI” by definition, which means the notice and governance duties aren’t hypothetical for us — they’re operational.
The Two Rules That Are Not on Grace Period
The grace period matters, so it’s worth being precise about what it actually covers: for most obligations under the Act, the government has said it will hold off on formal fact-finding and fines for at least a year from the January 22, 2026 effective date. Companies get time to build out risk-management systems, document their high-impact AI, and get processes in order before regulators come knocking.
Two things sit outside that window. First, failing to give advance notice that a product or service uses AI — a duty under Article 31(1) — is fined starting now, ₩5 million for a first offense. Second, failing to designate a domestic representative (국내대리인, gungnae daeriin, “domestic agent”) when one is legally required, under Article 36(1), is fined ₩20 million. Neither of these waits for 2027.
| Rule | Legal basis | First-offense fine | On grace period? |
|---|---|---|---|
| No advance AI notice to users | Article 31(1) | ₩5 million (≈ $3,700) | No — enforced now |
| No domestic representative | Article 36(1) | ₩20 million (≈ $14,900) | No — enforced now |
The domestic representative rule is aimed at foreign AI operators, not Korean banks like mine, but it’s worth knowing the trigger since a lot of the tools we use at work — from cloud AI services to third-party HR platforms — are built by companies headquartered outside Korea. A foreign operator has to appoint a Korean representative if it hits any one of three thresholds, and it only takes one.
| Threshold (any one triggers the duty) | Value |
|---|---|
| Prior-year total company revenue | Over ₩1 trillion (≈ $745 million) |
| Prior-year AI-service revenue specifically | Over ₩10 billion (≈ $7.5 million) |
| Average daily domestic users, past 3 months | Over 1 million users/day |
Read those numbers next to the ₩20 million fine and the incentive is obvious: a mid-size foreign AI vendor operating in Korea without a domestic representative is now gambling with a fine that resets, not a one-time slap.
What This Looks Like From an Employee’s Desk
None of this is abstract to me anymore, because I’ve started seeing it. Our internal loan-screening dashboard now shows a small notice banner the first time you open a case file in a given session — plain language stating that an AI model contributed to the risk score, with a link to the methodology summary. It wasn’t there in December. Our recruiting portal added something similar: candidates now see a line during the application process disclosing that resume screening is AI-assisted, which is the advance-notice duty in Article 31(1) showing up exactly where the law says it should — hiring.
I’ve written before about what it’s actually like working IT at a Korean financial company — the internal systems, the compliance culture, the way regulation lands on engineers before it lands on the public. This is the first time I’ve watched a specific legal deadline turn into a UI change I could point to on my own screen within a matter of weeks, rather than a policy memo that sits in a compliance folder nobody opens.
Our performance-evaluation software is next in line for the same treatment, according to a note from HR — evaluation algorithms are explicitly employment-adjacent high-impact use cases, so if your Korean employer uses any kind of automated scoring for reviews, expect a similar banner or disclosure to show up there too, grace period or not.
A Checklist for Foreign Workers in Korea
If you work at a Korean company in finance, healthcare, hiring/HR tech, energy, or transportation, here’s what’s actually worth doing right now rather than waiting for 2027:
- Look for a notice banner the next time you use an internal tool that scores, ranks, or screens something — loan applications, resumes, performance reviews. If AI is involved and there’s no notice, that’s the exact gap Article 31(1) targets.
- Ask HR or IT directly whether your company has classified any of its internal systems as “high-impact AI” (고영향 인공지능). Most compliance teams will have an answer already, even if it hasn’t been communicated widely.
- Check whether hiring tools you interact with disclose AI use — as a candidate or as someone on a hiring panel. This is one of the few places the law gives individual workers something concrete to point to.
- Don’t assume “grace period” means “nothing enforced.” I made that mistake once already. Two specific violations carry fines today, not in 2027.
- If you’re using AI tools from a foreign vendor at work — HR software, evaluation platforms, anything cloud-based — it’s fair to ask whether that vendor has a Korean domestic representative if it’s a large-enough operation. Your compliance team will likely already be asking the same question.
FAQ
Does the grace period mean I have no protections right now?
No. Most administrative fact-finding and fines are deferred for at least a year from January 22, 2026, but advance AI notice failures (Article 31(1), ₩5 million) and missing domestic representatives (Article 36(1), ₩20 million) are enforced immediately.
What counts as “advance AI notice”?
A plain-language disclosure, given before or at the point of use, that a product, service, or decision involves AI — the kind of banner I now see on our loan-screening and recruiting tools at work.
Who actually needs a domestic representative?
Foreign AI operators that meet at least one of three thresholds: over ₩1 trillion in prior-year total revenue, over ₩10 billion in AI-service revenue specifically, or an average of more than 1 million domestic daily users over the preceding three months.
What should I do if my company hasn’t posted any AI notice at all?
Ask HR or IT compliance whether the relevant system has been classified as high-impact AI. If it scores, ranks, or screens people — hiring, credit, performance — it likely qualifies, and the notice duty already applies regardless of the grace period.
Where This Leaves Me
The correction I owe my coworker is a small thing — a follow-up message, an “actually, two things don’t wait” — but it changed how I read every other piece of Korean regulation that gets described in headlines as “delayed” or “phased in.” A grace period is never a blanket, and the parts that get carved out first tend to be the parts regulators consider non-negotiable: transparency and accountability, not the deeper technical compliance work that genuinely takes companies a year to build. If you work anywhere near hiring, credit, healthcare, or evaluation systems in Korea, the banner you haven’t noticed yet is worth five minutes of actually reading.